FASTA Toolkit

Files are read with fastatacular.read_fasta, which also parses UniProt header fields (OS=, OX=, GN=, PE=, SV=) and pipe-delimited identifiers such as sp|P12345|NAME_HUMAN. Compression is detected from the first bytes of each file.

Entries whose identifier starts with a common decoy prefix (DECOY_, decoy_, REV_, rev_, XXX_, SHUFFLED_, RANDOM_ and their lower-case forms) are counted as existing decoys.

Each entry is tested in this order and dropped at the first filter it fails: header regular expression, organism, minimum length, maximum length, allowed residues. The organism filter compares the text after OS= (case-insensitive substring) and the number after OX=.

Duplicates are then removed, keeping the first copy: by identifier, then by sequence. With "Treat I and L as the same", sequences that differ only by I/L count as duplicates, because they have the same mass and give the same peptide masses.

Length is the number of residues per entry; composition counts the 20 standard residues over all entries, and other characters are listed separately. Organisms come from the UniProt OS= and OX= header fields.

"Compute stats and digest" cuts every sequence with the selected protease and counts peptides within the length range. "Peptides" counts every peptide of every protein; "distinct peptides" counts each sequence once. Protease rules are the cleavage regular expressions in tacular's protease table, the same ones peptacular.digest uses.

Decoys are made with fastatacular.make_decoys. Each decoy keeps its target's header fields; only the identifier gets the prefix.

• reverse: the sequence reversed. pseudo-reverse: each stretch between kept residues (K and R by default) reversed, so tryptic peptide lengths and masses match the targets. Both are deterministic (Elias and Gygi 2007).
• shuffle: residues randomly permuted; redrawn up to 10 times if it gives back the target.
• de Bruijn: every distinct (k+1)-mer in the database gets one replacement residue, so sequence repeated between targets is repeated between decoys (Moosa et al. 2020).
• Markov chain: residues sampled from a chain trained on a proteome, using the previous order residues as context.

"Decoy = target" counts decoys identical to their target. "Composition difference" is the sum over residues of |target fraction − decoy fraction| (0 = identical composition, 2 = no residue in common).

Targets and decoys are digested in silico with the selected protease, missed cleavages and length range. T is the set of distinct target peptides and D the set of distinct decoy peptides.

• Shared decoy peptides = |T ∩ D| / |D|: the fraction of distinct decoy peptides that are also target peptides. A search engine reports those as target hits, so they cannot count as decoy hits.
• Shared if I = L: the same, after replacing I with L in both sets (isoleucine and leucine have the same mass).
• Decoy / target peptides = |D| / |T|. Target-decoy FDR estimation assumes about 1.

Masses are neutral monoisotopic peptide masses (residue masses plus water) from peptacular's residue table. Changing only the decoys reuses the target digest.

FASTA is written with fastatacular.write_fasta. CSV has one row per entry with the columns of fastatacular.to_records (identifier, parsed UniProt fields, sequence and length).

PEFF 1.0 is written with pefftacular.write_peff, one database block per identifier prefix (sp, tr, …). Blocks whose prefix starts with a decoy prefix (the one set on the Decoys tab or a common one) are marked Decoy=true. The file names of the loaded files go in DbSource.